JPH0425738B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION Technical Field of the Invention The present invention relates to a method of PCM encoding and transmitting an input signal based on the difference between an input signal and a linear prediction value for the input signal, and in particular, a method for transmitting a PCM encoded input signal using a difference between an input signal and a linear prediction value for the input signal. The present invention relates to an adaptive predictor that can ensure the stability of a filter constituting the adaptive predictor that adaptively varies according to a signal.

Prior Art and Problems When encoding and transmitting highly correlated signals such as audio signals, there is a limited number of ways to encode and transmit signals based on the difference between the input signal and the linear predicted value for the input signal. It is known to be effective for transmission rates.

FIG. 1 shows the configuration of a differential PCM encoder and decoder. In the figure, a indicates an encoder, and b indicates a decoder.

In FIG. 1a, 1 is a subtracter which outputs a difference value between an input signal sampled at a constant period and a predicted value. A quantizer 2 quantizes the difference value and outputs a quantized error signal.
This signal is a desired PCM signal, and is transmitted via a transmission path and is input to an adder 3 where it is added to the predicted value. The output of adder 3 is a decoded value approximately equal to the input signal. 4 is a predictor,
The next sampling value is predicted from the decoded value based on an arbitrary prediction function for each sampling period and output as a predicted value.

In FIG. 1b, numeral 5 is an adder that adds the input error signal and the predicted value to generate and output a decoded value. A predictor 6 predicts the next sampling value from the decoded value based on the same prediction function as in the predictor 4, and outputs it as a predicted value.

In the encoding system shown in Figure 1,
Generally, the higher the prediction order in the predictors 4 and 6, the higher the prediction accuracy, and even higher prediction effects can be obtained by adaptively changing the prediction coefficients in the predictors according to the signal. However, when making the prediction coefficient variable, the stability of the predictor that constitutes the filter as a filter generally becomes a problem. For this reason, adaptive predictors that adaptively vary coefficients are conventionally implemented using first- or second-order prediction coefficients, and do not use third-order or higher order prediction coefficients, in order to easily control the stability of the filter. figure,
Therefore, it was not possible to obtain a sufficient predictive effect.

Purpose of the Invention The present invention aims to solve the problems of the prior art, and its purpose is to evaluate the stability of the operation of an adaptive predictor having a third-order pole by using a relatively simple relational expression. The purpose of this invention is to provide an adaptive predictor guaranteed by

Structure of the Invention The adaptive predictor of the present invention includes a linear prediction filter having a third-order pole in a transfer function, and adaptively updates the tap coefficients of each coefficient unit sequentially. After updating the coefficients, the third
The tap coefficients given by the first and second coefficient machines can take the roots of the polynomial denominator of the transfer function of the prediction filter within the unit circle on the Z plane using the tap coefficient value given by the coefficient machine as a parameter. A range is determined and the first and second tap coefficient values are modified so that they fall within this range.

Embodiment of the Invention FIG. 2 shows the configuration of an embodiment of the adaptive predictor of the present invention, and illustrates the configuration when applied to a decoder. In the figure, the same numbers as in Figure 1 indicate the same parts, and 11, 12, and 13 are each 1
Sampling period delay circuit, 14, 15, 16
are prediction coefficient units, and 17 is an adder.

The transfer function of the linear predictive filter constituting the adaptive predictor shown in FIG. 2 is expressed by the following equation.

1/(1- _a1Z - ¹ _- a2Z ^-2 -a3Z ^- ₃ )...(1) Here, _a1 to _a3 are respective prediction coefficients. The condition for a filter with such a transfer function to operate stably is that when the roots of the following equation Z ³ −a ₁ Z ² −a ₂ Z−a ₃ = 0 are found, The value is 1 or less. However, in a signal processing device that operates online, it is impossible to solve a cubic equation such as equation (2) and obtain _the absolute value every time. It is necessary to establish limiting conditions that give the range of _a3 .

In the adaptive predictor of the present invention, as a coefficient limiting method that can be easily realized in a signal processing device, when coefficient a ₃ is used as a parameter, other coefficients
A method is used to ensure the stability of the filter by giving the possible ranges of a ₁ and a ₂ .

Now, on the condition that the absolute values of all roots of equation (2) are less than or equal to 1, the range of coefficients that guarantees the stability of the filter can be numerically determined. Note that the filter operates stably when the absolute value of the root of the polynomial in the denominator of the transfer function expressed by equation (2) is 1 or less, that is, when it is within the unit circle on the Z plane, for example, (L, Rabiner, R.
W. Schafer; Translated by Hisaki Suzuki, "Digital Signal Processing of Audio," published by Corona Publishing, pp. 20-22).

Figure 3 shows the range in which the absolute value _{of the root of equation (2) is less than or equal to 1 when a 3 is} fixed to a certain constant value in equation ( ₂ ) and the roots of equation (2) are found while changing the combination of a 1 and a 2. This shows the results obtained numerically using a computer. In the figure, the horizontal axis shows the coefficient a ₁ and the vertical axis shows the coefficient a _2. When the coefficient a ₃ is used as a parameter, a ₂ = 1 + a ₁ + a ₃ a ₂ = 1 - a ₁ - a ₃ a ₂ as shown in the figure. = _a32-1 - _a1 ^- _a1 , a range A surrounded by _a3 indicates a stable range.

In addition, as a result of the above computer analysis, when the absolute value of a ₃ is 1 or more, no combination of a ₁ and a ₂ for which the absolute value of the root of equation (2) is 1 or less can be found, so the filter |a ₃ |<1...(3) is obtained as a conditional expression for each coefficient used to ensure the stability of. The relational expression between a ₁ and a ₂ showing the stability region in Fig. 3 is: a ₂ > a ₃ ² -1-a ₁ , a ₃ a ₂ <1+a ₃ +a ₁ a ₂ <1-a ₃ -a ₁ Become.

Considering the influence of the finite word length on the signal processing device, it is necessary to limit the coefficients to the inside of the stable region shown in FIG. Therefore, using a positive constant ε smaller than 1, the following conditional expression is obtained.

a ₂ ＞(a ₃ ² −1)ε−a ₁ , a ₃ ...(4) a ₂ ＜(1+a ₃ )ε+a ₁ ...(5) a ₂ ＜(1−a ₃ )ε−a ₁ ... ...(6) In (4) to (6), ε shows the effect of restricting the coordinates a ₁ and a ₂ in FIG. 3 to positions closer to the origin as ε approaches 0.

FIG. 4 shows the configuration of an embodiment of the adaptive predictor of the present invention, and shows a specific example of the configuration of each coefficient unit 14, 15, 16 of the prediction filter in the adaptive predictor shown in FIG. There is. The figure shows a case in which each coefficient of the prediction filter is obtained by successively correcting the ME simple method, which is a type of minimum error type adaptive equalization algorithm, and 21, 22, 23 are correlators,
4, 25, and 26 are integrating circuits, 27, 28, and 29 are limiting circuits, and 30, 31, and 32 are delay circuits.

In FIG. 4, the correlators 21, 22, 23 output the decoded value y(n) and the output signals y(n-1), y(n-2), y(n-3) of the delay circuits 11, 12, 13. After calculating the polarity correlation of each of them, an output A _i (n+1) multiplied by a constant α is output. That is, Ai (n + 1) = αsgn (x (n) sgn (y (n - i)) i = 1, 2, 3 In addition, in Fig. 4, the integration circuits 24, 25, 26
performs an integral operation as shown in the following equation.

a _i ′(n+1)=ra _i (n)+A _i (n+1) i=1, 2, 3 where r is a leakage constant. Furthermore, the conditions expressed by equations (3) to (6) for each integrator output a _i '(n+1) are realized by limiting circuits 27, 28, and 29. In other words, in the limiting circuit 27, a ₁ (n+1)= (a ₃ (n+1)-2)ε where a ₁ (n+1)≦(a ₃ (n+1)-2)
When ε.

(a ₃ (n+1)+2)ε.

However, a ₁ (n+1)≧(a ₃ (n+1)+2)
When ε.

a ₁ '(n+1) In cases other than the above.

In the limiting circuit 28, a ₁ (n+1) = (a ₃ (n+1) ² -1)ε-a ₁ (n+1) a ₃
(n+1) where a ₂ (n+1)≦(a ₃ (n+1) ² −1)ε
-a ₁ (n+1) a ₃ (n+1).

(1+a ₃ (n+1))ε+a ₁ (n+1) However, a ₂ (n+1)≧(1+a ₃ (n+1))ε
+a ₁ (n+1).

(1-a ₃ (n+1)) ε-a ₁ (n+1) where a ₂ (n+1) ≧ (1-a ₃ (n+1)) ε
-a ₁ (n+1).

a ₂ ′(n+1) In cases other than the above.

In the limiting circuit 29, a ₃ (n+1)=1 However, when a ₃ (n+1)≧1.

-1 However, when a ₃ (n+1)≦-1.

a ₃ '(n+1) In cases other than the above.

shall be.

With such a circuit configuration, it is possible to update the prediction coefficient of the third-order pole while ensuring stability.

Effects of the Invention As explained above, according to the adaptive predictor of the present invention, the root of the polynomial in the denominator of the transfer function of the linear prediction filter is set within the unit circle using the tap coefficient value given by the third coefficient unit as a parameter. As shown in FIG. After updating the tap coefficients of the coefficient multiplier, the possible range of each tap coefficient is determined and each tap coefficient is modified. This improves the stability of the operation in an adaptive predictor equipped with a linear prediction filter having a third-order pole. can be secured.

[Brief explanation of drawings]

FIG. 1 is a diagram showing the configuration of a differential PCM encoder and decoder, FIG. 2 is a diagram showing the configuration of an embodiment in which the adaptive predictor of the present invention is applied to a decoder, and FIG.
The figure shows a range in which stability of operation is guaranteed in a prediction filter having a third-order pole. FIG. 4 is a specific configuration example of each coefficient unit of the prediction filter in an embodiment of the adaptive predictor of the present invention. FIG. 1...Subtractor, 2...Quantizer (Q), 3...
Adder, 4... Predictor (P), 5... Adder, 6
...Predictor (P), 11,12,13...Delay circuit (Z ^-1 ), 14,15,16...Prediction coefficient unit, 1
7... Adder, 21, 22, 23... Correlator, 2
4, 25, 26...Integrator circuit, 27, 28, 29
...Limiting circuit, 30, 31, 32...Delay circuit (Z ^-1 ).

Claims (1)

[Claims] 1 Linear prediction having a third-order pole in a transfer function, comprising three stages of delay circuits connected in series in series and first, second, and third coefficient multipliers provided at the output of each delay circuit. In an adaptive predictor that has a filter and adaptively updates the tap coefficients provided by each coefficient unit for each input sample according to a minimum error type adaptive equalization algorithm, the tap coefficients provided by the third coefficient unit are applied. means for determining the possible range of the tap coefficients provided by the first and second coefficient multipliers so that the root of the polynomial of the denominator of the transfer function exists within a unit circle on the Z plane using the tap coefficient value given by the coefficient unit as a parameter; , means for correcting the first and second tap coefficient values so that they fall within the range, and after updating the tap coefficients of each coefficient unit, determining the possible range of each tap coefficient and correcting each tap coefficient. An adaptive predictor characterized by: